Gemini Omni Flash GA: extension and interpolation vs Sume's inputs

Google lists Gemini Omni Flash as generally available with video extension, interpolation between images and resolutions from 360p to 4K. Sume's catalog entry gemini-omni-flash-1.1 documents text, image, reference and video-to-video inputs at 3 to 10 seconds, and its docs do not list an extension feature, so an extension workflow on Sume means generating a new clip rather than continuing one.

The two lists describe different layers, so compare them as capability checklists, not as one product.

What each side lists

The Google column is as listed on the Gemini API changelog. The Sume column is from the Video Router docs.

How Sume routes by input shape

gemini-omni-flash-1.1 is one catalog id, and Sume picks the capability from what you send. You never choose an endpoint. The mapping is documented as follows.

The closest thing to interpolation

Sending image_url with end_image_url is the documented start and end frame input. Whether it matches what Google calls interpolation between images is not something Sume's docs claim, so test it on your own frames and judge the motion yourself before relying on it.

Native audio is always on, so generate_audio: false is rejected, and there is no bitrate_mode and no reference_audio_urls.

Working around the 10 second limit

If you need a longer shot, plan it as several clips.

• Generate the first clip, extract a frame near its end with a frame tool, and use it as the image_url of the next clip.
• Or write each clip as its own shot and join them on a timeline.
• For a longer single clip, a different catalog model may fit: seedance-2.5 accepts 4 to 30 seconds. Read capabilities from GET /v1/video-router/models rather than assuming one envelope.

A request

This is a start-and-end-frame request on the Video Router. Billing for this model is provider list times 1.25 per output second by resolution, so preview with dry_run through MCP or check the pricing page before a large batch.

curl -X POST https://api.sume.com/v1/video-router/generate \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: omni-frames-001" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "A slow dolly move from the first frame to the last",
    "image_url": "https://example.com/start.jpg",
    "end_image_url": "https://example.com/end.jpg",
    "resolution": "720p",
    "mode": "async"
  }'

Choosing between the two for your workflow

If your product needs a continuous shot longer than ten seconds from one prompt, check each provider's current page for the exact extension behavior, because the changelog line is short and the details live in Google's own documentation. If your product needs a controlled start and end look, the Sume input shape is explicit and easy to test: two images, one prompt, one request, and the same envelope as every other Video Router call.

Either way, run a small test with your own frames before you commit a batch. Motion between two very different images can look abrupt. Keep the first test at 720p to hold cost down, and move to 1080p or 4K only for clips you intend to ship.

A note on editing

video_url is the edit source, not a reference, and cannot be combined with image_url, end_image_url or reference_*_urls. If your goal is to change part of an existing clip, that is the supported path, and it is a separate capability from extension.